Photon-Counting CT Reconstruction With a Learned Forward Operator

Photon-Counting CT is an emerging imaging technology that promises higher spatial resolution and the possibility for material decomposition in reconstruction. A major difficulty to efficiently model cross-talk between detectors. In this work, we accelerate image reconstruction tasks by modelling with appropriately trained deep convolutional neural network. The main result relates proving convergence when using such a learned context of second-order optimisation methods spectral CT. Another evaluate method through numerical experiments on small-scale acquisitions generated realistic physics model. Using full as ground truth, results 20 dB increase peak-signal-to noise ratio compared ignoring altogether. At same time, it effectively cuts computation time half. Furthermore, generalises well both unseen data detector settings. Our indicate partially forward operator suitable way generation computational benefit becomes more noticeable problem sizes.